Exascale Computing Project Driving a HUGE Change in a ... · Exascale Computing Project – Driving a HUGE Change in a Changing World Al Geist Chief Technology Officer Exascale Computing

www.ExascaleProject.org

Exascale Computing Project –Driving a HUGE Change in a Changing World

Al Geist Chief Technology Officer

Exascale Computing Project

OpenFabrics Conference

March 27, 2017

Austin, TX

http://www.exascaleproject.org/

2 Exascale Computing Project, www.exascaleproject.org

Leverages the planned DOE facilities acquisitions◊ 2017 CORAL (collaboration of ORNL, ANL, LLNL)◊ 2020 APEX (collaboration of LANL/SNL, NERSC)◊ 2022 CORAL◊ 2024 APEX

Exascale Computing Project - Lift the entire HPC ecosystem and enable continued U.S. leadership in HPC

Time (CY)

Capability

2017 2021 2022 2023 2024 2025 2026 2027

10X

5X


Reaching the Elevated Trajectory will require solving key exascale challenges

• Extreme Parallelism– For example, an Exaflop @ 1 GHz requires a billion threads executing

• Memory and Storage– BW, latency, and capacity are not scaling with flops

• Reliability– Energy saving techniques and number of components drive MTBF down

• Energy Consumption– 20MW per Exaflop has been a target since 2009

In addition, the exascale advanced architecture will need to solve

emerging data science and machine learning problems in addition

to the traditional modeling and simulations applications.


Radical, Novel, Advanced solutions are not a Requirement but may be needed

We want the vendors to propose what they see as being needed to meet performance, reliability, programmability, data science convergence, and power requirements.

• If vendors can meet the requirements without needing new radical solutions that is fine and likely preferred.

• If it involves radical new concepts, we are interested in hearing about these solutions.

• We want to encourage vendors to propose new ideas where they provide a path for addressing our requirements but we don’t need novelty or “advancedness” just so we can claim things are “advanced”.


Goals of the Exascale Computing Project

Develop scientific, engineering, and large-data applications that exploit the emerging, exascale-era computational trends caused by the end of Dennard scaling and Moore’s law

Foster application development

Create software that makes exascalesystems usable by a wide variety of scientists and engineers across a range of applications

Ease of use

Enable exascale by 2021 and by 2023 at least two diverse computing platforms with up to 50× more computational capability than today’s 20 PF systems, within a similar size, cost, and power footprint

Rich exascaleecosystem

Help ensure continued U.S. leadership in architecture, software and applications to support scientific discovery, energy assurance, stockpile stewardship, and nonproliferation programs and policies

US HPCleadership


The ECP Plan

• Use a holistic/co-design approach across four focus areas:

– Application Development

– Software Technology R&D

– Hardware Technology R&D

– Exascale Systems Development

• Enable an initial exascale system to be delivered in 2021 (power consumption and reliability requirements may be relaxed)

• Enable capable exascale systems to be delivered in 2022 as part of the CORAL DOE facility upgrades

• System acquisitions and costs are outside of the ECP plan, and will be carried out by DOE facilities


ECP Timeline has Two Phases – and ends 2022

R&D before facilities select exascale systems

Targeted development for known exascale architectures

2016 2017 2018 2019 2020 2021 2022 2023 20252024FY 2026

Exascale System #1Site Prep #1

Testbeds

Hardware Technology

Software Technology

Application Development

Facilities activities

outside ECP

NRE System #1

NRE System #2

Exascale System #2Site Prep #2


What about the 2021 System?

• The site of the 2021 system is TBD and will be decided by DOE around June 2017.

• If the site is one of the CORAL labs, then the CORAL RFP will state:

“Within the goal of having three capable exascale systems by 2022-2023, if an early exascale system can be delivered in 2021 and upgraded to a capable exascale system by 2023, then provide the upgrade as an option.”

• If the site of the 2021 system is outside the CORAL labs then (in addition to the CORAL RFP for three 2022 systems) a separate RFP for a single 2021 system will be released in 2018 by the chosen lab.


What is a capable exascale computing system?

ECP defines a capable exascale system as a supercomputer that

• Can solve science problems 50x faster (or more complex, for example, more physics, higher fidelity) than the 20 PF systems of today can solve comparable problems.

• Must use a software stack that meets the needs of a broad spectrum of applications and workloads

• Have a power envelope of 20-30 MW

• Must be sufficiently resilient such that user intervention due to hardware or system faults is required on the order of a week.


Diversity is Very Important to DOE

• In 2018 a single CORAL RFP will be released for delivery of three capable exascale systems by the 2022-2023 timeframe. The RFP will also include NRE for the systems.

• These systems will have to be designed to solve emerging data science and machine learning problems in addition to the traditional modeling and simulations applications.

• The DOE Leadership Computing Facility has a requirement that the ANL and ORNL systems must have diverse architectures.

• Given the ECP goal of fostering a rich exascale ecosystem, LLNL has the option to choose a system that is diverse from both the ANL and ORNL systems.


There are Many Types of System Diversity

• Systems can vary from one another in many different dimensions

– System (architecture, interconnect, IO subsystem, density, resilience, etc.)

– Node (heterogeneous, homogeneous, memory and processor architectures, etc.)

– Software (HPC stack, OS, IO, file system, prog environment, admin tools, etc.)

– Hardware e.g.

• Ways Systems can be diverse

– Few big differences

– Many little differences

– Different technologies

– Different ecosystems, i.e., vendors involved

technology

scaletype

DDR NV PIM

sizeDIMM

Memory

on diestacked

Fat thin accel

#coreshomo

Processor

hetero

topologies

perfoptical

Network

copper


How Diverse is Enough?

How diverse is enough? There is no hard metric, Labs will evaluate diversity by how much it will benefit the exascale ecosystem

Having system diversity provides many advantages.

• It promotes price competition, which increases the value to DOE.

• It promotes a competition of ideas and technologies, which helps provide more capable systems for DOE’s mission needs.

• It reduces risk that may be caused by delays or failure of a particular technology or shifts in vendor business focus, staff or financial health.

• It helps promote a rich and healthy high performance computing ecosystem, which is important for national competitiveness and DOE’s strategic plan.


The ECP holistic approach uses co-design and integration to achieve capable exascale

Application Development

SoftwareTechnology

HardwareTechnology

ExascaleSystems

Scalable and productive software

stack

Science and mission applications

Hardware technology elements

Integrated exascalesupercomputers

Correctness Visualization Data Analysis

Applications Co-Design

Programming models, development environment,

and runtimesTools

Math libraries and Frameworks

System Software, resource management threading, scheduling, monitoring, and control

Memory and Burst

buffer

Data management I/O and file

systemNode OS, runtimes

Resili

ence

Work

flow

s

Hardware interface

Co-design centers

Proxy apps

Integration of NNSA

and Office of Science

SW efforts

PathForward

Design Space Evaluation

Testbeds

NRE


Application Scope determined by Mission Needs

• Materials discovery and design

• Climate science

• Nuclear energy

• Combustion science

• Large-data applications

• Fusion energy

• National security

• Additive manufacturing

• Many others!

• Stockpile Stewardship Annual Assessment and Significant Finding Investigations

• Robust uncertainty quantification (UQ) techniques in support of lifetime extension programs

• Understanding evolving nuclear threats posed by adversaries and in developing policies to mitigate these threats

• Discover and characterize next-generation materials

• Systematically understand and improve chemical processes

• Analyze the extremely large datasets resulting from the next generation of particle physics experiments

• Extract knowledge from systems-biology studies of the microbiome

• Advance applied energytechnologies (e.g., whole-device models of plasma-based fusion systems)

Key science and technology challenges to be addressed

with exascale

Meet national security needs

Support DOE science and energy missions


ECP Application Development – (1/3)

Climate (BER)

Accurate regional impact assessment of

climate change*

Combustion (BES)

Design high-efficiency, low-

emission combustion engines and gas turbines*

Chemical Science (BES,

BER)

Biofuel catalysts

design; stress-resistant crops

Fundamental Laws (NP)

QCD-based elucidation of fundamental

laws of nature: Standard

Model validation and

beyond SM discoveries

MaterialsScience (BES)

Find, predict, and control

materials and properties:

Applications chosen based on National impact and DOE Offices priorities


ECP Application Development – (2/3)

Genomics (BES)

Protein structure and dynamics; 3D

molecular structure design of

engineering functional properties*

Precision Medicine for Cancer (NIH)

Accelerate and translate cancer

research in RAS pathways,

drug responses, and

treatment strategies*

Seismic(EERE, NE,

NNSA)

Reliable earthquake

hazard and risk assessment in

relevant frequency ranges*

treaty verification

assembled within the limitations of shared memory hardware, in addition to making feasible the assembly

of several thousand metagenomic samples of DOE relevance available at NCBI [40] .

Figure 1: NCBI Short Read Archive (SRA) and

HipMer capability growth over time, based on rough

orderofmagnitude estimates for 1% annual compute

allocation (terabases, log scale).

Figure 2. Current (green area) and projected (pink

area) scale of metagenomics data and

exascaleenabled analysis.

Furthermore, the need for efficient and scalable de novo metagenome sequencing and analysis will only

become greater as these datasets continue to grow both in volume and number, and will require exascale

level computational resources to handle the roughly doubling of metagenomic samples/experiments every

year and the increased size of the samples as the cost and throughput of the sequencing instruments

continue their exponential improvements. Increasingly it will be the genome of the rare organism that

blooms to perform an interesting function, like eating the oil from the Deep Water Horizon spill [41,42],

or provides clues to new pathways and/or diseases.

Assembling the genomes from hundreds of thousands of new organisms will provide us with billions of

novel proteins that will have no sequence similarity to the currently known proteins from isolate genomes.

The single most important method for understanding the functions of those proteins and studying their

role in their communities is comparative analysis, which relies on our ability to group them into clusters

of related sequences. While this is feasible for the proteome of all “isolate” genomes ( i.e. , from cultured

microorganisms; currently comprising around 50 million proteins), it is currently impossible for the

proteome of metagenomic data (currently at tens of billion proteins).

2.3 RELEVANT STAKEHOLDERS

This proposal supports directly the main two research divisions of DOE’s Biological and Environmental

Research (BER), namely the Biological Systems Science Division (BSSD) and the Climate and

Environmental Sciences Division (CESD). Furthermore, several other funding agencies have a strong

interest in microbiome research [40] . These include (a) federal agencies already funding largescale

metagenome sequencing or analysis projects, such as NIH (Human Microbiome Project), NSF

(EarthCube initiative), USDA, NASA, DoD; (b) philanthropic foundations such as the Gordon and

Betty Moore Foundation (Marine Microbiome Initiative), Simons Foundation, Bill and Melinda Gates

Foundation, Sloan foundation (indoor microbiome), etc.; (c) pharmaceutical industry such as Sanofi. In

addition, the workload represented by these applications are quite different than most modeling and

simulation workloads, with integer and pointerintensive computations that will stress networks and

5

Metagenomic (BER)

Leveraging microbial

diversity in metagenomicdatasets for

new products and life forms*

Chemical Science (BES)

Design catalysts for

conversion of cellulosic-

based chemicals into

fuels, bioproducts

Some applications also include a significant machine learning component *


ECP Applications Development – (3/3)

* Scope includes a significant data science component

Demystify origin of universe

and nuclear matter

in universe*

Astrophysics (NP)

Cosmology (HEP)

Cosmological probe of

standard model (SM) of particle

physics: Inflation, dark matter, dark

energy*

Magnetic Fusion

Energy (FES)

Predict and guide stable

ITER operational

performance with an

integrated whole device

model*

Nuclear Energy (NE)

Accelerate design and

commercialization of next-generation

small modular reactors*

Wind Energy (EERE)

Increase efficiency and reduce cost of turbine wind

plants sited in complex terrains*

Some applications also include a significant data science component *


ECP Application Development Co-Design Centers

• Center for Online Data Analysis and Reduction at the Exascale (CODAR)

• Block-Structured AMR Co-Design Center (AMReX)

• Center for Efficient Exascale Discretizations (CEED)

• Co-Design Center for Particle Applications (CoPA)

• Graph and Combinatorial Methods for Enabling Exascale Applications (GraphEx)


ECP Software Technology Summary

• ECP will build a comprehensive and coherent software stack that will enable application developers to productively write highly parallel applications that can portably target diverse exascale architectures

• ECP will accomplish this by extending current technologies to exascalewhere possible, performing R&D required to conceive of new approaches where necessary, coordinating with vendor efforts, and developing and deploying high-quality and robust software products


ECP Hardware Technology Summary

Objective: Fund R&D to design hardware that meets ECP’s Targets

for application performance, power efficiency, and resilience

• Issue PathForward and PathForward-II Hardware Architecture R&D contracts

• Participate in evaluation and review of PathForward and LeapForwarddeliverables

• Lead Design Space Evaluation through Architectural Analysis, and Abstract Machine Models of PathForward/PathForward-II designs for ECP’s holistic co-design


Goals for PathForward (issued last year – vendor awards pending)

• Improve the quality and number of competitive offeror responses to the

Capable Exascale Systems RFP

• Improve the offeror’s confidence in the value and feasibility of aggressive

advanced technology options that would be bid in response to the Capable

Exascale Systems RFP

• Improve DOE confidence in technology performance benefit,

programmability and ability to integrate into a credible system platform

acquisition


Goals of PathForward-II (planned for issue in 2017)

• Support high payoff, innovative hardware technologies and systems technologies that may have higher risk. It is focused on component, node, and system architecture designs that will intersect with the 2021 exascale system.

• Also of interest to the PathForward-II RFP team:

– Innovations that may enable dramatic acceleration of certain applications, for example, delivering a 100x increase in 2021 on some classes of applications while still being able to solve the full range of DOE applications

– Developments that promote wider diversity in the exascale ecosystem

– Innovations in power consumption, performance, programmability, reliability, data science, machine learning, or portability

– Reducing total cost of ownership


ECP Exascale Systems Summary

• Funds Non-Recurring Engineering (NRE)

– Brings to the product stage promising hardware and software research and integrates it into a system

– Includes application readiness R&D efforts

– Must start early enough to impact the system - more than two full years of lead time are necessary to maximize impact

• Funds Testbeds

– ECP ECP testbeds will be deployed each year throughout the project

– FY17 testbeds will be acquired through options on existing contracts at Argonne and ORNL

– Testbed architectures will track SC/NNSA system acquisitions and other promising architectures


This is a very exciting time for computing in the US

• Unique opportunity to do something HUGE for the nation in HPC

• The exascale systems in 2021 and 2022 afford the opportunity for

– More rapid advancement and scaling of mission and science applications

– More rapid advancement and scaling of an exascale software stack

– Rapid investments in vendor technologies and software needed for 2021 and 2022 systems

– More rapid progress in numerical methods and algorithms for advanced architectures

– Strong leveraging of and broader engagement with US computing capability

• When ECP ends, we will have

– Prepared industry and critical applications for a more diverse and sophisticated set of computing technologies, carrying US supercomputing well into the future

– Demonstrated integrated software stack components at exascale

– Invested in the engineering and development, and participated in acquisition and testing of capable exascale systems


Thank you!




Documents

Exascale Computing Project Driving a HUGE Change in a ... · Exascale Computing Project – Driving a HUGE Change in a Changing World Al Geist Chief Technology Officer Exascale Computing